Development of a Domain-Specific AI Job-Seeking Support Platform Based on Prompt Engineering: Focusing on the HAIJOB Case

Abstract

This study presents HAIJOB, a domain-specific AI job-seeking support platform that generates Korean-style self-introduction essays from users’ experience data. Prior AI recruitment-document technologies have mainly addressed resume information extraction and job description (JD) matching, leaving question-specific, narrative-based essay generation underexplored. HAIJOB addresses this gap through a six-layer architecture that embeds prompt engineering as an independent processing layer and integrates user experiences, target jobs, postings, and essay questions into a unified context for the LLM engine. Operational data from 56,722 essay generations and 21,885 revisions revealed two user groups: draft-centered users (64.8%) and iterative writers (35.2%). Across 6,160 essay–JD pairs, aligned pairs showed significantly higher similarity than random controls (Cohen’s d = 1.74). In 511 JD–interview question pairs, generated questions reflected an average of 35.8% of job-posting keywords. User feedback analysis further identified directions for improving output quality. This study contributes a transferable implementation model for domain-specific LLM services by extending recruitment-document generation from resume-centered matching to question-responsive essay generation.